Fast high-signal-to-noise ratio equivalent time processor

Abstract

The present invention generally relates to ultra high bandwidth, ultra high ample rate systems and more particularly to equivalent time sampling and signal reconstruction. The disclosed technique provides a means to obtain high SNR (signal to noise ratio), uniformly spaced, ET (equivalent time) measurements from data where each point may have differing SNR and/or where the data is not uniformly spaced. The invention disclosed provides least-square optimum SNR output data from a plurality of arbitrarily time-skewed input data. The invention further discloses an implementation of the technique that requires no matrix inversion, but instead puts nearly all operations in a vector multiply or vector add form. The disclosed technique vectorizes the required equations allowing the invention to be implemented in low cost real-time hardware--that is, DSP (digital signal processing) integrated circuits, common micro-processors, or special purpose chip sets.

Claims

1. An equivalent time processor, comprising:

A first input means for receiving a plurality of input data records wherein each input data record has a plurality of equally spaced data points representing a signal plus noise which has been sampled with a sample spacing of.tau., and wherein the plurality of input data records are time-skewed with respect to each other in an arbitrary fashion, relative to said signal;

A second input means for receiving a set of skew-time, T.sub.s, measurements of the time between a trigger synchronous with said signal and the sample-time of the first data point in each of said plurality of input data records;

A sorting means that directs said input data records in one of R ways according to whether their skew-time falls within R contiguous ranges, wherein R can be chosen arbitrarily;

A plurality of R accumulators which accumulate all records directed to them by said sorting means,

A plurality of R counting means which provide a plurality of counts Q.sub.n, of the number of records accumulated in each of said plurality of R accumulators, wherein n indexes the plurality of accumulators and is typically n=0,1,2,... R-1, to produce the plurality of counts, Q.sub.n;

A division means to compute the plurality of numbers q.sub.n =1/Q.sub.n for all said plurality of counts Q.sub.n provided by said plurality of R counting means;

A plurality of scalar-vector-multiplier means which multiply the accumulated results in each of said plurality of R accumulators, by its associated said q.sub.n provided by said division means, to produce a plurality of R averaged records;

A processing means that uses said set of skew-time measurements, said plurality of counts Q.sub.n, and said plurality of R averaged records to generate an output data record having a plurality of equally spaced data points representing said signal plus lower noise, which has been sampled with a sample spacing of m.tau./n where m and n are any positive integer, wherein said processing means combines said plurality of arbitrarily skewed input data records such that said plurality of samples in said output record have a least-squares optimal SNR (signal to noise ratio), and wherein said processing means optimally preserves frequency components beyond the Nyquist limit of said input data records by combining said plurality of arbitrarily skewed input data records such that said plurality of samples in said output record contain frequency components beyond the Nyquist limit of said input data records;

An output means for providing said output record as high SNR, high bandwidth, uniformly spaced, equivalent time measurements.

2. The process of claim 1 wherein said processing means comprises:

A coefficient generation algorithm that uses said plurality of counts Q.sub.n, to compute a plurality of coefficient vectors, wherein each vector is a set of coefficients that is optimal in a the weighted least squares sense, for fitting a curve through p points in said plurality of averaged records, surrounding in time, an output point, and wherein the weights used for the weighted least squares computation are provided by said counts Q.sub.n; so that the coefficients generated account for the arbitrary, and sometime zero, SNR of said averaged data records;

A plurality of coefficient applicator means which accepts said plurality of coefficient vectors and said plurality of R averaged records and produces a plurality of smoothed, high SNR data records;

An interleaver that interleaves the plurality of said smoothed, high SNR data records to produces a uniformly spaced, high SNR interleaved data record.

3. The process of claim 2 wherein said plurality of coefficient applicator means is implemented with fast vector-scalar-multiply hardware which is easily pipelined to carryout multiple operations simultaneously.

4. The process of claim 1 wherein said processing means comprises:

A coefficient generation algorithm that uses said plurality of counts Q.sub.n, to compute a plurality of coefficient vectors, wherein each vector is a set of coefficients that is optimal in a the weighted least squares sense, for fitting a curve through p points in said plurality of averaged records, surrounding in time, an output point, and wherein the weights used for the weighted least squares computation are provided by said counts Q.sub.n; so that the coefficients generated account for the arbitrary, and sometime zero, SNR of said averaged data records;

An interleaver the interleaves said plurality of averaged data records to produces an interleaved data record with non uniform SNR;

A plurality of adaptive finite impulse response (FIR) filters, each having a plurality of taps, wherein the tap weights are provided by said coefficient vectors generated by said coefficient generation algorithm;

A distribution network which feeds said interleaved data record to the plurality of FIR filters;

An interleaver that accepts the output of the plurality of FIR filters, decimates the output of each FIR filter as needed, and combines them outputs of said FIR filters and generate said output data record.

5. The process of claim 2 or 4, wherein said coefficient generation algorithm is implemented without any matrix inversions, but as a closed form solution--with weighting from said counts Q.sub.n as parameters--such that it is implemented with simple and fast vector addition and vector multiplication, and vector gather operations that are optimal for digital signal processing (DSP) hardware and digital logic.

6. The process of claim 4 wherein said FIR filters have decimated outputs to minimized the operations required to implement said FIR filters.